2019 Prognostics & System Health Management Conference—Qingdao
(PHM-2019 Qingdao)
A new method for estimating lithium-ion battery
capacity using genetic programming combined model
Hang Yao,Xiang Jia*,Bo Wang, Bo Guo
College of Systems Engineering
National University of Defense Technology
Changsha, China
jiaxiang09@sina.cn
Abstract—Lithium-ion battery is the main energy source widely
used in many fields. Therefore, it is particularly essential for
estimating the health of lithium-ion battery accurately, especially
in important fields such as aerospace, rail transit and satellite.
For lithium-ion battery, the battery capacity is a health index
(HI) that best reflects its performance degradation. By estimating
the battery capacity, the health status of the lithium-ion battery
can be clearly identified. However, there are technical barriers to
the direct measurement of battery capacity in engineering, and
many characteristics and capacities of lithium-ion batteries have
abrupt changes, so that it is difficult to calculate the battery
capacity accurately by formula calculation. In this paper, a new
method of genetic programming combined model is proposed,
which can calculate the capacity of lithium-ion battery by
formulating multiple monitored features with a certain precision.
Therefore, the functional relationship between multiple features
and HI is well measured, which lays a good foundation for the
subsequent life prediction of battery.
Keywords-genetic programming; lithium-ion battery; health
index; formula calculation
I. INTRODUCTION
Nowadays, product health management is an important part
of improving industrial production efficiency and safety. With
the advancement of monitoring technology, using monitoring
data to predict product life and achieve health management
has become a viable and effective method. Lithium-ion
batteries are a fundamental component that is widely used in
satellite, aerospace and rail transit[1,2]. In order to meet the
needs of operating voltage and power, as well as to extend the
life of the battery pack and battery, it is especially important to
achieve health management of lithium-ion batteries.
These fields have fully tested the performance status of
lithium-ions, and obtained characteristic data of many lithium-
ion batteries. Although the battery capacity can directly reflect
the health indexs of lithium-ion battery performance
degradation, due to the limitations of monitoring technology,
the battery capacity cannot be directly measured in actual
engineering. In order to solve this problem and realize further
management of lithium-ion batteries, many scholars have done
a lot of effective work on the state-of-health of lithium-ion
battery.
The features of lithium-ion battery such as the temperature,
current, voltage or resistance of the lithium-ion battery that are
used as HIs to predict the remaining life. For making full use
of the performance characteristics of the monitoring, more
accurate methods for judging the health status of lithium ions
have been applied. Landi and Gross [3] estimated the overall
trend of battery status based on the relationship between fuzzy
logic processing performance characteristics and battery state.
Liu [4] et al. use a combination of nonlinear degradation
functions to optimize the autoregressive model to achieve an
estimate of battery cycle life. Li [5] et al. used artificial neural
networks (ANNs) to predict Remaining Useful Life (RUL) for
lithium-ion batteries. Zhang [6] et al. adopt Accelerated
Particle Swarm Optimization (APSO) algorithm to optimize
the kernel function of Relevance Vector Machine (RVM) to
realize the prediction of lithium-ion battery RUL.
Currently, multiple features are mapped to battery capacity
by RVM or Support Vector Machine (SVM) methods to
achieve residual life prediction. SVM and RVM realize good
results in estimating the state-of-health of lithium-ion
batteries, and have obtained more in-depth research. Wang et
al. [7] established a three-parameter conditional capacity
degradation model and used RVM to obtain correlation
vectors to represent battery capacity attenuation and cycle life.
Simultaneously. Widodo et al. [8] proposed a battery health
assessment framework based on discharge voltage sample
entropy. Based on discharge voltage sample entropy, Widodo
et al. [8] proposed a battery health assessment framework and
gave the uncertainty representation. Li et al. [9] established a
multi-step prediction model to realize the state-of-health
prediction based on average entropy and RVM. Zhang et al.
[10] used wavelet to denoise the noise during battery testing,
and estimated the robustness of the test by differential
evolution-optimized RVM. These methods have good results
in the estimation of lithium-ion health state, but these methods
can not reflect the functional relationship between the selected
features and battery capacity. It is necessary to select the
appropriate kernel function in the estimation according to the
actual situation.
A method of genetic programming combined model is
proposed in this paper. According to the multiple
2019 Prognostics & System Health Management Conference—Qingdao (PHM-2019 Qingdao)
978-1-7281-0861-2/19/$31.00 ©2019 IEEE